CN109583095B - Extended Period Forecast Method of Northwest Pacific Typhoon Based on Hybrid Statistical Dynamic Model - Google Patents

Abstract

The invention relates to a North Pacific typhoon extension period forecasting method based on a hybrid statistical power model, which utilizes a modern statistical method to classify the track of historical typhoons, and establishes a statistical forecasting equation of intra-season oscillation signals and typhoons by the influence of intra-season scale sea gas states on different types of typhoons; and forecasting typhoons generated in 10-30 days in the future by using a high-resolution sea-air coupling power mode as a forecasting factor for the forecasting field of the oscillating signals in the seasons of 10-30 days in the future. And finally, multiplying the generation number of typhoons of different types by the climate probability distribution of the track of the typhoons, so that the probability distribution map of the generation and the frequency of the typhoons on the entire North Pacific ocean is predicted 10-30 days in advance. The beneficial effects are that: the total number of typhoons on the North and North Pacific ocean can be effectively predicted in the extension period of the typhoons, which is 10-30 days in the future, and the spatial probability distribution of typhoons generated/frequency in 10 days can be obtained by utilizing different generation positions and movement tracks of the typhoons.

Description

Extended Period Forecast Method of Northwest Pacific Typhoon Based on Hybrid Statistical Dynamic Model

technical field

The invention relates to the field of atmospheric science and technology, in particular to a method for forecasting extended periods of typhoons in the Northwest Pacific Ocean based on a hybrid statistical dynamic model.

Background technique

Geological disasters such as violent winds, heavy rain, huge waves, storm surges and indirect landslides and debris flows caused by typhoons often cause heavy casualties and social and economic losses. The Northwest Pacific is the sea area with the highest frequency of typhoons, accounting for about 30% of the number of tropical cyclones in the global sea area. After the typhoon is formed in the tropical western Pacific Ocean and the Philippine Sea, it develops and migrates west/northwest, and may invade the southeastern coastal areas of my country. Statistics show that between 1983 and 2006, an average of 6-7 typhoons landed in my country's coastal provinces every year, causing tens of billions of yuan/ Years of economic losses and thousands of casualties per year. Therefore, the research and development of key technologies for typhoon monitoring and forecasting has become a major demand for national disaster prevention and mitigation and social and economic policy formulation.

At present, the typhoon forecasting systems of major operational forecasting units in my country and the world include: 1) medium and short-term (within 5 days) high-resolution numerical model forecasting and 2) monthly-scale climate forecasting. Due to the noise of the atmosphere itself and the systematic error of the numerical model, the accuracy of the typhoon forecast for more than one week is low (Vitart et al. Scale dynamic process, insufficient ability to simulate typhoon intensity and path. Therefore, as a supplement to the numerical model, the monthly prediction of typhoon is usually based on the statistical prediction model (Gray 1984; Chan et al. 1998; Fan and Wang 2009) and the hybrid- Statistical models (Murakami et al. 2016).

There is an obvious gap between short-term weather forecasts and climate forecasts, and the development and improvement of extended-period (10-30 days) weather forecast models to complete a seamless forecast system is the primary task of global weather and climate forecast research (Waliser 2005) . The source of the predictability of the extended-range forecast mainly comes from the intraseasonal oscillation in the atmosphere (Madden and Julian 1994; Li Chongyin et al. 2003; Waliser 2005), and the tropical-subtropical intraseasonal oscillation has a significant impact on the Northwest Pacific typhoon (Ding Yihui et al. 1977; Gray 1979; Liebmann et al.1994; Maloney and Hartmann 2000; Zhu Congwen et al. 2004; Kim et al.2008; Sun Chang et al. 2009; Li Chongyin et al. 2012; Both the sexual circulation and the convergence zone are conducive to synoptic-scale disturbances gaining kinetic energy from intraseasonal oscillations, so more typhoons occur and intensify (Chen Guanghua and Huang Ronghui 2009; Hsu et al. 2011). Camargo et al. (2009) and Zhao et al. (2015) also pointed out that the mid-level water vapor field and low-level vorticity field on the intraseasonal scale are closely related to the low-frequency changes in typhoon activities. Correlation, forecast typhoons on an extended period scale.

Although previous studies have found the importance of intraseasonal oscillations on the occurrence and development of typhoon activities, the method of applying the correlation between the two to the extended period forecast of typhoon activities in the Northwest Pacific Ocean has not been established. The current literature only focuses on tropical cyclones in the southern hemisphere. Statistical forecast of weekly changes (Leroy and Wheeler 2008) and dynamic model evaluation (Vitart et al. It is possible to forecast the generation of typhoons 2–4 weeks in advance (Fuand Hsu 2011; Wu and Duan 2015; Xiang et al.2015). Therefore, the development of extended-range forecasting methods and models for the Northwest Pacific Ocean has high meteorological operational application value.

Contents of the invention

The purpose of the present invention is to overcome the above-mentioned deficiencies in the prior art and provide a method for forecasting the extended period of a typhoon in the Northwest Pacific based on a hybrid statistical dynamic model, which is specifically realized by the following technical solutions:

The Northwest Pacific typhoon extension period forecasting method based on the hybrid statistical dynamic model is based on the typhoon optimal path data set provided by the Joint Typhoon Warning Center JTWC under the Pacific Meteorological and Oceanographic Center of the US Navy, including the following steps:

Step 1) First, the tropical cyclone data of the JTWC tropical cyclone data set are divided into several categories according to the generation location and development trajectory by using the c-means fuzzy clustering analysis method;

Step 2) Find the statistical relationship between the low-frequency large-scale field and the number of typhoons generated by various types, and establish a statistical forecasting equation for forecasting the number of typhoons generated by each type;

Step 3) Bring the corresponding forecast factors in the large-scale low-frequency field output by the GFDL model into the empirical historical statistical forecasting equation to obtain the number of typhoon anomalies per ten-day period during the forecast period from 2003 to 2012; add the number of typhoon anomalies to the typhoon The total number of typhoons predicted by the historical climate average number of seasons;

Step 4) Multiply the total number of each type of typhoon obtained by the forecast by the probability distribution of each type of typhoon's ten-day-by-ten-day climatic average generation position and trajectory distribution, and obtain the occurrence probability of each type of typhoon's generation position and trajectory in each ten-day period, respectively, After summing up the occurrence probabilities of the generation locations and trajectories of all typhoons, the probability distribution map of the generation locations and trajectory frequencies of typhoons over the entire Northwest Pacific Ocean in each ten-day period is obtained;

In said step 2), in order to establish a forecasting equation with relatively stable forecasting performance, four methods are used to define the predictor, which are respectively:

The first method of defining predictors: perform regional average in the box of the area where historical typhoons are most concentrated in each forecast field, and the box of the area described for each type of typhoon is fixed;

The second method of defining predictors: find the largest significant positive correlation area frame and the largest significant negative correlation area frame on the correlogram of the number of typhoon anomalies and the low-frequency large-scale field, and perform regional average respectively, for each For each large-scale environmental field of a typhoon, the size and position of the corresponding regional frame will change;

The third method of defining the predictor: In the historically largest area of typhoon generation, obtain the positive and negative grid point average of the number of typhoon anomalies and the correlation coefficient of the low-frequency sea-air field that passes the 95% significance test;

The fourth method of defining predictors: Find a large-scale area that passes the 95% significance test on the correlation map of the number of typhoon anomalies and the low-frequency large-scale field. If there are positive and negative correlation areas that meet the conditions at the same time, the two The ones are subtracted to synthesize a predictor;

In the step 2), after defining the predictors by the four methods to obtain the predictors of each type of typhoon, use the multiple linear stepwise regression method to establish empirical historical statistical forecasts for each type of typhoon predictors and the number of corresponding historical typhoon anomalies Equation; through the stepwise regression method, the predictors with the best forecasting performance and independent of each other are selected to avoid overfitting.

The further design of the Northwest Pacific typhoon extension-period forecasting method based on the hybrid statistical dynamic model is that the coefficient of the positive correlation grid point in the third method of defining the predictor is 1, and the coefficient of the negative correlation grid point is -1.

The further design of the Northwest Pacific typhoon extension period forecasting method based on the hybrid statistical dynamic model is that the generation position and trajectory of each type of typhoon in the step 4) in each ten-day probability are as follows:

in, Indicates the climate probability of typhoons of type C _k generated or passed in the 5°×5° box of the latitude and longitude grid (i, j) within the first ten days of the climate average, Indicates the frequency of generation or passage of typhoons of type C _k in the 5°×5° box of the latitude and longitude grid (i, j) within the first ten days of the climate average, Indicates the total frequency of typhoons of type C _k generated or passed over the entire Northwest Pacific within the lth tenth day of the climate average.

The further design of the Northwest Pacific typhoon extension period forecasting method based on the hybrid statistical dynamic model is that the generation position and track occurrence probability of each type of typhoon in each ten-day forecast are as shown in formula (2):

in, Indicates the probability that the forecast typhoon of type C _k will generate or pass within the 5°×5° box of the latitude and longitude grid (i,j) within the lth tenth day of the year p; Indicates the number of typhoons of type C _k generated within the lth tenth day of year p, and N _{Fcsttotal,l,p} indicates the total number of typhoons generated in the entire Northwest Pacific within the lth tenth day of year p; is the average climate probability of typhoons of category C _k in formula (1).

The further design of the Northwest Pacific typhoon extension period forecast method based on the hybrid statistical dynamic model is to sum up the probabilities of each type of typhoon to obtain the total probability of typhoon generation or passage on the latitude and longitude grid (i, j) as shown in the formula ( 3)

Among them, nk is the total number of typhoons in c-means cluster analysis, and nk is 7 for typhoons in the northwest Pacific Ocean.

The further design of the Northwest Pacific typhoon extension-period forecasting method based on the hybrid statistical dynamic model is that step 1) also includes a preprocessing operation on the return data of the model: setting the forecast object as the generation of typhoons in the Northwest Pacific every ten days Number, generation location and trajectory probability distribution map; the modeling object of this method is set as May 16-December 5 in 1979-2002, and the annual modeling time date corresponds to the forecast time time date; And it is set to forecast the number of typhoons per ten-day anomaly after removing the climate average seasonal variation. The climate average typhoon seasonal cycle component corresponding to the seasonal variation is averaged every ten days during the 24-year modeling period of 1979-2002.

The further design of the Northwest Pacific typhoon extension period forecasting method based on the hybrid statistical dynamic model is that the preprocessing operation also includes: extracting the low-frequency components in the large-scale field of the model return data, specifically including the following steps:

Step A) Remove the annual cycle and the first three harmonics from the daily large-scale field to obtain a new field;

Step B) Subtract the moving average of the previous 120 days from the field obtained in step A) to obtain the anomaly field with the interannual variation removed;

Step C) The anomaly field is averaged for another 10 days according to the date of the forecast period to obtain the low-frequency large-scale environmental field corresponding to the typhoon anomaly data every ten days.

The advantages of the present invention are as follows:

The prediction results of the four methods for finding potential key predictors in the present invention are all very close, as shown in Fig. 5, which illustrates the stability of the method. The result of 0 days in advance (Lead0) is obtained by bringing the observation data in the forecast period into the forecast equation, and the forecast result can be regarded as the forecast upper limit of the forecast model. The forecast results at 10 days in advance are very close to the upper limit of the model forecast (that is, the result of Lead 0), and the correlation coefficient reaches 0.45–0.46. As the forecast lead time increases, the forecast skill begins to decline. The coefficient skill is also 0.21, passing the 95% reliability test.

It can be seen from Figure 5 that the forecast results of the four methods for selecting predictors are similar, and Figure 6 shows the forecast results of the simple ensemble average of the four methods. The time series of typhoon anomalies forecasted 0 days in advance (part a in Figure 6) shows that the correlation coefficient skill of the upper limit of the model's forecast is 0.52, and when the forecast is 10 days in advance, the forecast skill of the model is still as high as 0.46 (the Part b), as the forecast time advances, the correlation coefficient keeps decreasing, and the root mean square error keeps increasing. Taking the 95% significance test as the standard of forecasting skills, the Northwest Pacific typhoon forecasting method based on the hybrid statistical dynamic model is The effective forecasting ability of the number of typhoon anomalies is 15–20 days. If the total number of typhoons is obtained by adding the forecasted typhoon anomalies to the climate average seasonal cycle component, see Figure 7, the 30-day advance forecast still shows the height forecast skill.

Description of drawings

Figure 1 is a flow chart of the dynamic-statistical forecasting model for typhoon forecasting in the Northwest Pacific over the extended period.

Figure 2 is a schematic diagram of the trajectories and average trajectories of seven types of typhoons in the JTWC typhoon dataset from June to November 1979-2002 obtained by using the c-means fuzzy clustering analysis method.

Figure 3 is a schematic diagram of the power spectrum analysis of the ten-day anomalies of seven types of typhoons in the JTWC typhoon dataset from June to November in 1979-2002.

Figure 4 shows the time correlation coefficient diagram of the annual low-frequency OLR and VWS anomalies and the number of typhoon C1 decadal anomalies from June to October in 1979-2002 and the regional schematic diagram of four selected key predictors.

Figure 5 is a schematic diagram of the evaluation of the forecast results of the four methods for selecting key predictors. Among them, part a-d of Figure 5 is a schematic diagram of the time correlation coefficient of the forecast results of the four methods of selecting key predictors; part e-h of Figure 5 is a schematic diagram of the root mean square error of the forecast results of the four methods of selecting key predictors ; Part i–l of Figure 5 is a schematic diagram of the AUC index of the forecast results of the four methods for selecting key predictors.

Figure 6 shows the observation and advance (Part a of Figure 6) 0 days, (Part b of Figure 6) 10 days, (Part c of Figure 6) 15 days from May 16 to December 5, 2003-2012, ( Figure 6 part d) 20 days, (Figure 6 part e) 25 days and (Figure 6 part f) 30 days forecast time of mixed forecast typhoon anomaly data of TCall and C1–C7 sum.

Figure 7 shows the correlation between the number of typhoon anomalies (part a-c of Figure 7) and the total number (part d-f of Figure 7) of mixed forecasts and observations (part a of Figure 7, part d of Figure 7) Coefficients, (Part b of Figure 7, Part e of Figure 7) root mean square error and (Part c of Figure 7, Part f of Figure 7) probability forecasting AUC index.

Fig. 8 shows the observation probabilities (parts a–d of Fig. 8) of the four cases of typhoon generation, the probability of “perfect reconstruction” (parts e–h of Fig. 8), and (parts i–l of Fig. 8) 0 days ahead, Schematic diagram of the forecast results for 10 days in advance (part m–p of Figure 8) and 15 days in advance (part q–t of Figure 8).

Figure 9 shows the observation probabilities (parts a–d of Figure 9) of the four typhoon frequency cases, the “perfect reconstruction” probability (parts i–l of Figure 9) of 0 days in advance, (Part m–p of Fig. 9) 10 days in advance and (part q–t of Fig. 9) 15 days in advance forecast results.

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings.

As shown in Figure 1, the Northwest Pacific typhoon forecast method based on the hybrid statistical dynamic model provided in this embodiment is based on the typhoon optimal path data set provided by the Joint Typhoon Warning Center JTWC under the Pacific Meteorological and Oceanographic Center of the US Navy, including the following steps :

Step 1) Firstly, the tropical cyclone data of JTWC tropical cyclone data set from May to November in 1979-2002 were divided into seven categories by c-means fuzzy clustering analysis method according to the generation location and development trajectory. The classification results are shown in Figure 2. Each type of tropical cyclone has a unique generation location and development track, and the number of ten-day typhoons of each type has a significant 10–90-day subseasonal variation feature, see Figure 3.

Typhoon Best Track Dataset: Provided by the Joint Typhoon Warning Center (JTWC) under the Pacific Meteorological and Oceanographic Center of the US Navy. The data time length is from 1979 to 2012, and the time interval is 6 hours. According to the definition of Saffir-Simpson's tropical cyclone level, the forecast object of this method is the tropical storm with the maximum surface wind speed greater than 34knots.

Step 2) Find the statistical relationship between the low-frequency large-scale field and the number of typhoons generated by various types, and establish a statistical forecasting equation for forecasting the number of typhoons generated by each type;

Step 3) Bring the corresponding forecast factors in the large-scale low-frequency field output by the GFDL model into the empirical historical statistical forecasting equation to obtain the number of typhoon anomalies per ten-day period during the forecast period from 2003 to 2012; add the number of typhoon anomalies to the typhoon The total number of typhoons predicted by the historical climate average number of seasonal changes.

The US GFDL high-resolution air-sea coupling model has 32 vertical layers in the atmospheric model with a horizontal resolution of 50 km, and 50 vertical layers in the ocean model with a horizontal resolution of 1°×1°. The climate model outputs a forecast result on the 1st, 6th, 11th, 16th, 21st and 26th of each month from April to November in 2003-2012, and the results are displayed at 00:00, 04:00, and 08:00 UTC every day At 12:00 and 16:00 as different initial conditions, there are 5 forecast sets in total, and each forecast set is integrated 50 days later. Therefore, there are 2400 return data (10 years x 8 months x 6 days x 5 collections) in the 10 years. The large-scale sea-air field predicted by this method using the model includes: Outgoing Longwave Radiation (OLR), surface temperature (T _s ), specific humidity at 700hPa, vertical velocity at 500hPa, vertical wind shear, divergence and vorticity at 850hPa Fields, a total of 7 thermal and dynamic large-scale environmental fields related to typhoon generation.

Step 4) Multiply the total number of each type of typhoon obtained by the forecast by the probability distribution of the climate average generation position and track distribution of each type of typhoon every ten days, and obtain the generation position and track occurrence probability of each type of typhoon in each ten-day period , after summing up the probabilities of generation locations and trajectories of all typhoons, the probability distribution map of generation locations and track frequencies of typhoons over the entire Northwest Pacific Ocean in each ten-day period is obtained.

In the above step 2), in order to establish a forecasting equation with relatively stable forecasting performance, take the correlation coefficient diagram between the number of C1 typhoon anomalies and OLR and vertical wind shear as an example, see Figure 4, and introduce the four methods used in this embodiment. One way to define predictors:

The first method of defining predictors: take the regional average of the area where historical typhoon generation is most concentrated in each forecast field (the box on part a of Figure 4 and e in Figure 4). Therefore, the frame of this area is fixed for each type of typhoon, that is, the same frame is used for different large-scale fields of each type of typhoon, and each type of typhoon has 7 predictors.

The second method of defining predictors: find the largest significant positive correlation area frame and the largest significant negative correlation area frame on the correlogram of the number of typhoon anomalies and the low-frequency large-scale field, and perform regional average respectively (as shown in Figure 4 Part b of Figure 4 and part f of Figure 4). For each large-scale environmental field of each type of typhoon, the size and position of the box will vary. Therefore, each type of typhoon has 14 predictors.

The third method of defining the predictor: in the historically largest area of this type of typhoon generation (determined by the extreme latitude and longitude of the historical typhoon generation location, as shown in part c of Figure 4 and part g of Figure 4), find the number of typhoon anomalies The correlation coefficient with the low-frequency sea-air field passes the average of the positive and negative grid points of the 95% significance test, in which the coefficient of the positive correlation grid point is 1, and the coefficient of the negative correlation grid point is -1. There are 7 predictors for each type of typhoon in this method.

The fourth method of defining predictors: looking for a large-scale area that passes the 95% significance test on the correlogram of the number of typhoon anomalies and the low-frequency large-scale field (see part d of Figure 4 and part h of Figure 4). If there are positive and negative correlation areas that meet the conditions at the same time, the two are subtracted to synthesize a predictor. Therefore, there are seven predictors for each type of typhoon in this method.

In the above step 2), after defining the predictor by the above four methods to obtain the predictor of each type of typhoon, use the multiple linear stepwise regression method to establish an empirical historical statistical forecasting equation for each type of typhoon predictor and the number of corresponding historical typhoon anomalies; The above-mentioned stepwise regression method is used to select the predictors with the best forecasting performance and independent of each other to avoid overfitting.

The probability distribution of Northwest Pacific typhoons in the above step 4) is as follows:

in, Indicates the climate probability of typhoons of type C _k generated or passed in the 5°×5° box of the latitude and longitude grid (i, j) within the first ten days of the climate average, Indicates the frequency of generation or passage of typhoons of type C _k in the 5°×5° box of the latitude and longitude grid (i, j) within the first ten days of the climate average, Indicates the total frequency of typhoons of type C _k generated or passed over the entire Northwest Pacific within the lth tenth day of the climate average.

In the above step 4), the probability distribution of typhoon forecast C _k type is as formula (2):

in, Indicates the probability that the forecast typhoon of type C _k will generate or pass within the 5°×5° box of the latitude and longitude grid (i,j) within the lth tenth day of the year p; Indicates the number of typhoons of type C _k generated within the lth tenth day of year p, and N _{Fcsttotal,l,p} indicates the total number of typhoons generated in the entire Northwest Pacific within the lth tenth day of year p; is the average climate probability of typhoons of category C _k in formula (1).

After summing up the probabilities of each type of typhoon, the total probability of typhoons forming or passing over the entire Northwest Pacific Ocean is as shown in formula (3)

Among them, nk is the total number of typhoons in c-means cluster analysis, and nk is 7 for typhoons in the northwest Pacific Ocean.

As shown in Figure 8, the forecast results of four typhoon generation cases are shown. The spatial correlation coefficients of the forecast results 10 days in advance are all greater than 0.5, and the main areas of typhoon generation can be predicted. The probability of “perfect reconstruction” in parts e–h of Figure 8 represents the probability distribution of typhoons obtained by substituting the number of observed typhoons into Equation (2). The upper right corner is the spatial correlation coefficient between each forecast result and the observed field. Figure 9 shows the forecast results of four typhoon frequency (trajectory) cases. It can be seen from the forecast probability distribution trajectory map that these four cases have well predicted the main trajectory of typhoon movement in 10 days , the spatial correlation coefficient reached 0.6–0.8 at 10 days in advance. The "perfect reconstruction" probability in part e–h of Fig. 9 represents the typhoon probability distribution obtained by substituting the number of observed typhoons into Equation (2). The upper right corner is the spatial correlation coefficient between each forecast result and the observed field. This result well illustrates that the method of the present invention can not only effectively forecast the number of typhoons in the extended period, but also the total number of typhoons in the Northwest Pacific Ocean every ten days in the next 10-30 days, and can also use the different typhoons of these seven types The spatial probability distribution of typhoon generation/frequency within 10 days is obtained from the generation position and movement trajectory of typhoon.

In step 1), preprocess the return data of the model: set the forecast object as the number of typhoons generated in each ten-day period in the Northwest Pacific Ocean, the generation location and the probability distribution map of the trajectory; set the modeling object of this method as 1979–2002 From May 16th to December 5th, the annual modeling time corresponds to the forecast time; and the forecast is set to predict the number of typhoons per ten days after the seasonal variation of the climate average is removed. The climate mean typhoon seasonal cycle component corresponding to the change is averaged every ten days during the 24-year modeling period 1979-2002.

The above preprocessing operation also includes: extracting the intra-seasonal low-frequency components of the large-scale field of the model return data, specifically including the following steps:

Step A) Subtract the annual cycle and the first three harmonics from the daily large-scale field to obtain a new field.

Step B) Subtract the moving average of the previous 120 days from the field obtained in step A) to obtain the anomaly field with the interannual variation removed.

Step C) The above-mentioned anomaly field is averaged for 10 days according to the date of the above-mentioned forecast time, and the low-frequency large-scale environmental field corresponding to the typhoon anomaly data is obtained.

In this embodiment, after classifying the trajectories of historical typhoons using modern statistical methods, the statistical forecasting equations for intraseasonal oscillation signals and typhoon generation are established based on the influence of the air-sea state on the intraseasonal scale on the generation of different types of typhoons. Using the dynamical model to forecast the seasonal oscillation signal in the next 10 to 30 days as a forecast factor, after substituting it into the statistical equation, the generation of typhoons in the next 10 to 30 days can be forecasted. Finally, by multiplying the number of typhoons of different types by the climate probability distribution of their trajectories, the probability distribution map of typhoon generation and frequency over the entire northwest Pacific Ocean can be forecasted 10 to 30 days in advance.

In actual forecasting, steps 1) and 2) only need to be calculated once using historical observation data to obtain the generation and frequency probability maps of typhoons of different trajectory types, as well as the number of typhoons generated in history. Using the correlation between the number of typhoons generated in the historical data and the low-frequency and large-scale air-sea state analysis, an empirical forecasting equation for each type of typhoon is established. In the real-time forecasting operation, it is only necessary to bring the low-frequency and large-scale forecasting factors predicted by the dynamic model into the empirical forecasting equation in the next 10-30 days to obtain the number of anomalies generated by various typhoons in the next 10-30 days. It is multiplied by the climate average probability distribution map of various typhoons and summed up to obtain the probability distribution map of typhoons in the Northwest Pacific Ocean in the next 10 to 30 days.

The hybrid statistical dynamical model used in this method uses non-bandpass filtering method to extract the components of the intraseasonal scale (10–90 days) during the establishment process, so it can be directly applied to real-time forecasting. The result of the hybrid statistical dynamic model is the number of typhoons of each type and the total number of typhoons generated in the Northwest Pacific Ocean every ten days in the next 10 to 30 days, as well as the spatial distribution of the generation and frequency probability of typhoons in the Northwest Pacific Ocean.

The above is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed in the present invention can easily think of changes or Replacement should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be determined by the protection scope of the claims.

Claims (7)

1. A typhoon extension period forecast method in the Northwest Pacific Ocean based on a hybrid statistical dynamic model, based on the typhoon best path data set provided by the Joint Typhoon Warning Center JTWC under the Pacific Meteorological and Oceanographic Center of the U.S. Navy, characterized in that it comprises the following steps: Step 1) First, the tropical cyclone data of the JTWC tropical cyclone data set are divided into several categories according to the generation location and development trajectory by using the c-means fuzzy clustering analysis method; Step 2) Find the statistical relationship between the low-frequency large-scale field and the number of typhoons generated by various types, and establish a statistical forecasting equation for forecasting the number of typhoons generated by each type; Step 3) Bring the corresponding forecast factors in the large-scale low-frequency field output by the GFDL model into the empirical historical statistical forecasting equation to obtain the number of typhoon anomalies per ten-day period during the forecast period from 2003 to 2012; add the number of typhoon anomalies to the typhoon The total number of typhoons predicted by the historical climate average number of seasons; Step 4) Multiply the total number of each type of typhoon obtained by the forecast by the probability distribution of each type of typhoon's ten-day-by-ten-day climatic average generation position and trajectory distribution, and obtain the occurrence probability of each type of typhoon's generation position and trajectory in each ten-day period, respectively, After summing up the occurrence probabilities of the generation locations and trajectories of all typhoons, the probability distribution map of the generation locations and trajectory frequencies of typhoons over the entire Northwest Pacific Ocean in each ten-day period is obtained; In said step 2), in order to establish a forecasting equation with relatively stable forecasting performance, four methods are used to define the predictor, which are respectively: The first method of defining predictors: perform regional average in the box of the area where historical typhoons are most concentrated in each forecast field, and the box of the area described for each type of typhoon is fixed; The second method of defining predictors: find the largest significant positive correlation area frame and the largest significant negative correlation area frame on the correlogram of the number of typhoon anomalies and the low-frequency large-scale field, and perform regional average respectively, for each For each large-scale environmental field of a typhoon, the size and position of the corresponding regional frame will change; The third method of defining the predictor: In the historically largest area of typhoon generation, obtain the positive and negative grid point average of the number of typhoon anomalies and the correlation coefficient of the low-frequency sea-air field that passes the 95% significance test; The fourth method of defining predictors: Find a large-scale area that passes the 95% significance test on the correlation map of the number of typhoon anomalies and the low-frequency large-scale field. If there are positive and negative correlation areas that meet the conditions at the same time, the two The ones are subtracted to synthesize a predictor; In the step 2), after defining the predictors by the four methods to obtain the predictors of each type of typhoon, use the multiple linear stepwise regression method to establish empirical historical statistical forecasts for each type of typhoon predictors and the number of corresponding historical typhoon anomalies Equation; through the stepwise regression method, the predictors with the best forecasting performance and independent of each other are selected to avoid overfitting. 2. the Northwest Pacific typhoon extension period forecasting method based on the hybrid statistical dynamic model according to claim 1, is characterized in that the coefficient of the positively correlated grid point is 1 in the third kind of definition predictor method, and the coefficient of the negatively correlated grid point is- 1. 3. The Northwest Pacific typhoon extension period forecasting method based on the hybrid statistical dynamic model according to claim 1, characterized in that in the step 4), the probability of occurrence of each type of typhoon's generation position and track in each ten-day period is as follows: 1):

in,

Indicates the climate probability of typhoons of type C _k generated or passed in the 5°×5° box of the latitude and longitude grid (i, j) within the first ten days of the climate average,

Indicates the frequency of generation or passage of typhoons of type C _k in the 5°×5° box of the latitude and longitude grid (i, j) within the first ten days of the climate average,

Indicates the total frequency of typhoons of type C _k generated or passed over the entire Northwest Pacific within the lth tenth day of the climate average. 4. The Northwest Pacific typhoon extension period forecasting method based on the hybrid statistical dynamic model according to claim 3, characterized in that the generation position and track occurrence probability of each type of typhoon in each ten-day forecast are as formula (2):

in,

Indicates the probability that the forecast typhoon of type C _k will generate or pass within the 5°×5° box of the latitude and longitude grid (i,j) within the lth tenth day of the year p;

Indicates the number of typhoons of type C _k generated within the lth tenth day of year p, and N _{Fcsttotal,l,p} indicates the total number of typhoons generated in the entire Northwest Pacific within the lth tenth day of year p;

is the average climate probability of typhoons of category C _k in formula (1). 5. The Northwest Pacific typhoon extension period forecasting method based on the hybrid statistical dynamic model according to claim 4, characterized in that after summing up the probabilities of each type of typhoon, the generation or passage of typhoons on the latitude and longitude grid (i, j) is obtained The total probability of is as formula (3)

Among them, nk is the total number of typhoons in c-means cluster analysis, and nk is 7 for typhoons in the northwest Pacific Ocean. 6. The Northwest Pacific typhoon extension period forecasting method based on the hybrid statistical dynamic model according to claim 4, characterized in that step 1) also includes the preprocessing operation of the return data of the model: the setting forecast object is the Northwest Pacific The probability distribution map of the number of typhoons generated in each ten-day period, the generated location, and the trajectory; the modeling object of this method is set as May 16-December 5 in 1979-2002, and the annual modeling time date and forecast The times and dates correspond to each other; and it is set to forecast the number of typhoons per ten days after the average seasonal variation of the climate is removed. Get it on average every ten days. 7. The Northwest Pacific typhoon extension period forecast method based on the hybrid statistical dynamic model according to claim 6, characterized in that the preprocessing operation also includes: extracting the low-frequency components in the large-scale field of the model return data, specifically including Follow the steps below: Step A) Remove the annual cycle and the first three harmonics from the daily large-scale field to obtain a new field; Step B) Subtract the moving average of the previous 120 days from the field obtained in step A) to obtain the anomaly field with the interannual variation removed; Step C) The anomaly field is averaged for another 10 days according to the date of the forecast period to obtain the low-frequency large-scale environmental field corresponding to the typhoon anomaly data every ten days.

Images